AI wrapper app
An AI wrapper app is a program that doesn't develop the artificial intelligence itself, but rather uses that of a large provider over the internet and builds its own interface around it. The term is often used pejoratively because such providers are easily copied and depend on the provider's pricing.
Large language models like ChatGPT are programs that run on huge data centers and generate text. Developing them costs hundreds of millions of euros. That’s why most companies don’t build something like this themselves. Instead, they rent access: they send their users' requests over the internet to a large provider, get the answer back, and display it in their own app. Programs like this are called AI wrapper apps, from the English word “wrapper” for packaging. The actual thinking work happens elsewhere, the app is the shell around it.
Why “wrapper” is almost always meant as a criticism
In the tech industry, the word is a judgment, not a neutral technical term. Calling a company a wrapper means saying: there’s nothing of its own in there. The accusation targets two weaknesses. First, the app can easily be copied, because any competitor can tap into the same AI provider. Second, the business depends on a single supplier.
This dependency is real and has been tested multiple times. If the provider raises its prices, the wrapper app’s costs immediately rise along with them. If the provider builds the same feature directly into its own product, the reason to pay for the app disappears. Exactly this happened in 2023 to several small providers who made PDF documents searchable. When OpenAI built this feature directly into ChatGPT, their business collapsed. In the industry, this is called “getting steamrolled by the platform operator.”
But there is also a counterargument, and it should be taken seriously. Almost every successful piece of software builds on someone else’s technology. Spotify didn’t compose any music, Booking.com doesn’t own any hotels. What matters is whether the company adds something that is hard to replicate. Anyone who just puts a text field in front of the model is vulnerable. Anyone who brings their own data, vetted workflows, or established customer relationships is considerably less so.
The path of a request through the shell
Technically, this works via an interface, called an API for short. This is a fixed address on the internet to which you send a request and from which you get a response back. The wrapper app takes in the user’s input and prepends an invisible instruction to it in the background. This instruction is called a prompt and determines how the model should respond, for example as a matter-of-fact legal advisor or as a terse summarizer.
Additional building blocks are often added. Many apps first search their own documents and attach the relevant passages to the prompt. This process is called RAG and ensures that the model works with company-specific knowledge instead of just what it learned during training. Other apps check the response afterward, filter out nonsense, or ask a second time in unclear cases.
Payment is made per text chunk, referred to in technical jargon as a token. A token is roughly half a word to a whole word. So every request costs money, and it costs money anew every single time. Anyone selling a wrapper app for a monthly price therefore has to calculate carefully. Heavy users can make a subscription unprofitable.
Wrappers in the app store and in startup coverage
A large portion of the AI apps on your phone are wrappers. This includes translators, chatbots with personality, tools for writing job applications, or summarizers for long texts. You can often recognize them by the fact that they don’t work at all without an internet connection and name a third-party provider in their terms of use.
In business news, the term usually comes up during funding rounds. Investors then ask whether a startup has a so-called moat, meaning protection against imitators. “That’s just a wrapper” is a killer argument in such conversations. Conversely, it’s also true: companies like Cursor or Perplexity were called this early on and still became worth billions.
A common misconception is that wrappers are technically undemanding. That’s not true. Keeping costs under control, making responses reliable, and complying with data protection rules is demanding work. The difference doesn’t lie in the difficulty, but in who owns the most expensive component.